DelfinGroup is a Latvian fintech company that offers accessible, simple and consumer focused finance and pawn services.
Big data at Delfingroup
DelfinGroup uses big data carefully collected over 11 years of operation with hundreds of thousands of customers and millions of customer interactions and transactions.
Predictive models are built in-house on historical data by applying machine learning algorithms to identify patterns and trends in data that can be used to make better business decisions. Machine-learning algorithms are built in-house using the most popular programming languages, such as R and Python. DelfinGroup's data analysts use a wide variety of statistical and business intelligence software, including Power BI, Jasp, KNIME, Anaconda and others. DelfinGroup's business intelligence systems utilise data from cloud-based data lakes.
ETL tools are combined with Microsoft Power BI visualisation capabilities to visualise and present data in online reports available anytime and anywhere. Dashboards with carefully chosen KPIs are used to monitor and evaluate the performance of main business functions and departments. Access to real-time and near real-time visually immersive and interactive insights allows management to carry out quick and confident data-based decisions.
The application scoring engine has been developed using a combined stack of supervised classification algorithms such as random forests, decision trees and logistic regression. Statistical models are built on 100,000+ data points, including data provided by the customers when filling out a loan application, data on internal credit history and credit behaviour, as well as data from external sources such as credit bureaus, debt collection companies, the State Revenue Service and other publicly available databases. Scoring results provide not only a binary answer on whether to approve the application, but also assess the maximum credit risk and the maximum loan amount for each application. Fraud prevention is carried out by automatically flagging atypical customer behaviour and analysing phone, email, and location data which are especially important for online customers. The debt collection process is semi-automated, with machine-learning models assessing the probability of non-collection and subsequently modifying collection communication. Collection cases are assigned to collection specialists based on their historical performance. Marketing and sales communication is based on historical patterns, showing how various customer profiles react to different communication channels and messages. Communication strategies are constantly adjusted using the latest data from marketing and sales campaigns.